Online accelarated implementation of the Fuzzy C-means algorithm with the use of the GPU platform

Abstract

Fuzzy C-means is a very widely covered topic in literature. It is a very successful clustering method whose subtle variations are involved in various clustering related applications. Despite its success, it shares a disadvantage with almost all of its contemporary pattern discovery algorithms — computational complexity. With the explosion in multimedia data over the internet and growing storage systems, there is a lot of research done in content based data retrieval. Fuzzy C-means is an integral part of this goal but its innate complexity makes it a strictly offline algorithm. Online pattern discovery is the need of the hour and our paper aims to address this issue without the use of powerful servers for implementing Fuzzy C-Means (FCM). We aim at accelerating the algorithm using Graphical Processing Units (GPUs), which are basically graphic cards common in desktop computers. We aim at restructuring the algorithm in a manner in which maximum data parallelism could be extracted thus utilizing the resources of the GPU to the fullest extent. In this paper we compare the speed of our approach using a NVIDIA Tesla C1060 GPU to that of sequential versions running on an Intel Xeon 2.93 GHz and an Intel Dual Core 2GHz.